Glossary
World foundation model
A world foundation model is a foundation model trained to understand, generate, predict, or simulate environments. The term is especially common in physical AI, robotics, and synthetic data workflows.
Common contextPhysical AI
ExampleNVIDIA Cosmos
Used forSimulation, prediction, synthetic data
RelatedWorld model
Definition
A world foundation model is a large pretrained model for environments. It may reason about video, actions, motion, sound, future states, or physical interactions.
How it differs from creator world generation
Creator world generators often output scenes for humans to view. World foundation models for physical AI may output synthetic data, action predictions, or future-state estimates for robots and agents.
FAQ
Is every world model a world foundation model?
No. A world foundation model usually refers to a broader pretrained system that can support multiple downstream tasks, especially in physical AI.